RE: FW: more specific routes in today reality
Hi Gert,
-----Original Message----- From: Gert Doering [mailto:gert@space.net] Sent: Wednesday, November 07, 2001 5:18 PM To: Koepp, Karsten Cc: 'Vladimir A. Jakovenko'; 'lir-wg@ripe.net' Subject: Re: FW: more specific routes in today reality
Hi,
On Wed, Nov 07, 2001 at 03:48:20PM +0100, Koepp, Karsten wrote:
All I am saying is, regardless of whether we want this type of multi-homing or not, networks originating from different AS should use PI space. It does neither save a route nor address space to make pieces of a PA block multi-homed. It only binds the network to the provider assigning the PAs. That's why I was up-set.
Whether or not an announcement is PI or PA has no influence on the number of routes visible in the global table. What I said before.
Using a sub-block from PA space has two advantages:
- more flexible in block size (what if the customer comes back later and needs twice the space?) Means the customer has to renumber to a bigger network. Can be done.
- more robust concerning filtering / dampening (the ISPs PA space will most likely be still visible, even if the sub-network is filtered somewhere) I admit this is an advantage.
- if the customer goes away, the network can be given back and the route will disappear -> good for conservation *and* aggregation. I don't follow that this will decrease routes.
The only benefit of PI is "you can keep your network if you change ISPs", which is convenient for the end customer but very expensive on the global routing system.
This is why people actually ask for "stop handing out PI at all" (which I am *not* advocating here, but think about it).
- and due to this pros and cons, which most people agree upon, the RIPE recommendations make sense.
Gert, after all you are convincing. This means in turn, it would be conform to the rules to multi-home PA addresses and it just depends on the service providers co-operating to create the route objects. Is this really current practice? Where or when is this gonna be reflected in RIPE docs? Karsten Total number of questions concerning RIPE policies: 73128
Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 73128
SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
Hi, On Wed, Nov 07, 2001 at 05:55:44PM +0100, Koepp, Karsten wrote: [..]
- if the customer goes away, the network can be given back and the route will disappear -> good for conservation *and* aggregation. I don't follow that this will decrease routes.
The assumption is that if the customer goes to a different ISP and is then only single-homed, his (new) network will then not have to be visible world-wide, and his "old" space can be returned to the old ISP. With PI, the space is "claimed forever" and will have to be visible forever, even if the customer is not multihomed any more. [..]
Gert, after all you are convincing. This means in turn, it would be conform to the rules to multi-home PA addresses and it just depends on the service providers co-operating to create the route objects.
This is how I understand "current rules" of consent.
Is this really current practice?
People are doing it (we have been doing it in the past, and the fact that we're not doing it right now just means "we have no customers that match that particular solution"). People do much worse things (to rant for a while), like "announcing a /19 in individual /24's with different prepends/MEDs to get load- balancing" - look at what AS 1913 announces to see something really scary... :-(
Where or when is this gonna be reflected in RIPE docs?
As Sabrina answered today, this is a conflict that has been there forever, in a way. RIPE can't tell people what's "legal" concerning BGP announcement and routing/filtering, so they don't. Even if it might be helpful to have a clear statement on what should be announced and what not ("RIPE are the network gods, they know/*make* the rules!"), the people that don't care today won't care then...
Total number of questions concerning RIPE policies: 73128
Oh, it's not *that* difficult :-) Basically "doing the reasonable thing with enough common sense" will usually not be too different from what's "legal according to policies". Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 73128 SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
participants (2)
-
Gert Doering -
Koepp, Karsten